What Data for This Category Looks Like
Documents covered in this category include control plans, inspection records, SPC reports, finished product inspection reports, nonconforming product review forms, release forms, and quality monthly reports. Data sources include manufacturing execution systems (MES), online inspection equipment, laboratory systems, process parameter libraries, customer standard libraries, and historical defect libraries. Data is collected continuously in line with production batches and shifts. Significant differences exist between fields of different documents. Finished product inspection reports include quantitative fields such as batch pass rate and defect types. Nonconforming product review forms require association with process nodes and batch information.
Constraints on the Tool Calling and Plugins Workflow
Due to diverse document types and significant field differences, tool calling must adapt to the structured parsing logic of different documents to avoid cross-type data confusion. Data update frequency changes in real time with production batches. Tool calling must limit the batch range processed in a single call to prevent context overload. Multi-data source access requires plugins to support cross-system data pulling. Plugins must also verify field completeness to prevent incorrect releases caused by missing key information. The strictness of quality judgment requires tool calling to associate with the historical defect library to ensure accurate anomaly identification.
How to Configure Settings
This configuration applies to FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | Auto-match | Adapt to the parsing needs of multiple types of quality documents, covering different business scenarios such as control plans and inspection reports |
| Single-call data range | Single batch data | Match the feature that data is updated by production batch, avoid context overload |
| Context retention duration | 600 seconds | Match the average response cycle of the production process, prevent call timeouts |
| Abnormal field verification switch | Enabled | Prevent incorrect releases caused by missing qualified judgment fields, comply with quality control requirements |
| Workflow/API function-call configuration | Force verification of required fields | Ensure that the results returned by tool calls include core fields required for release, such as inspection pass rate and approval status |
Three Common Mistakes
- Custom plugins do not display input and output parameters in workflow tasks, and no error logs are generated. Cause: The parameter types and required attributes are not correctly declared in the plugin configuration, causing the system to fail to recognize the parameter structure.
- Looping through array elements does not trigger the expected AI conversation. Cause: Array elements are not correctly mapped to the input fields of the loop body, or the output format of the loop body does not meet the calling requirements of the AI conversation.
- Calling an external tool returns an error that VQD acquisition failed. Cause: The tool's access key or request parameters are not correctly configured, preventing a valid connection from being established with the third-party service.
How to Verify Successful Configuration
- Call a single control plan document, check whether the parsed results returned by the tool include the core fields of the document, and confirm that the matching logic of
toolChoicetakes effect. - Trigger a custom plugin test, check whether the workflow interface correctly displays the input and output parameters of the plugin, and confirm that the plugin configuration is correct.
- Simulate a call scenario with missing required fields, check whether the tool triggers field verification prompts, and confirm that the abnormal verification switch is working properly.
- Configure a loop run test case, check whether array elements are processed one by one, and confirm that the loop logic meets expectations.
The values given are common starting points and should be measured against the reader's own samples.